Atlas / Sociology of AI / Paper

A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms

Swarms & OversightIn-groups & MoralityNorms & Culture

Davide Paglieri, Logan Cross, Tim Genewein, Joel Z. Leibo, Nenad Tomašev, Alexander Sasha Vezhnevets · 2026 · arXiv (Google DeepMind)

Read the source

In a 100-agent swarm proving math conjectures, one agent’s grader exploit spread to the rest within half an hour. About a quarter of agents, unprompted, audited, warned, protested and boycotted — but had no power to act, so the authors call it a failure of institutions, not of norms.

Questions it bears on

Help improve this map

Know a paper that belongs here, or think something needs correcting?

Suggest a correction ↗Contributions are reviewed before they appear.